Abstract:

With the proposal of the "3060 Plan" and the introduction of a series of reform schemes for the power system, the development of new energy grid connection technology has been vigorously promoted. However, due to the randomness of photovoltaic power generation, accurately predicting photovoltaic power generation has become quite challenging. The large-scale connection of photovoltaic power plants to the power system poses severe challenges to the power system's power flow distribution and scheduling operations. This paper proposes an optimization scheduling method based on deep reinforcement learning, emphasizing its intelligence, self-regulation, and dynamic adjustment characteristics. It also attempts to explore multi-objective optimization and multi-level scheduling strategies, providing theoretical support and guidance for the efficient and sustainable development of energy storage systems.

Key words: deep reinforcement learning, energy storage system, energy management, optimization scheduling, multi-objective optimization